Apturo is a prospecting tool I built. It scrapes local businesses, audits their websites and Google listings, and generates a personalized report showing them exactly what they’re losing by missing calls and ranking poorly.

Then it sends that report out automatically. The whole point is to put outbound on autopilot so I’m not the bottleneck.

Except I was the bottleneck. Built it. Tested it. Rated it 8.5 out of 10. Then I just didn’t send the first 30 messages.
Not a tech problem. A me problem. Perfectionism dressed up as “still refining it.”

Here’s what changed it: I stopped using AI as an assistant and started using it as a sparring partner. Most people use Claude to execute tasks. I started asking it to find the thing I wasn’t seeing.

Four prompts that actually do this:

1. “After you finish what I asked, tell me one thing you’d do differently if this were your business. Even if I didn’t ask.”

By default, AI answers the question you ask and stops there. It won’t volunteer that your pricing is confusing or your offer is buried, even if it’s sitting right there in what you gave it. This prompt forces an opinion out of it instead of a deliverable. The trick is the “even if I didn’t ask” part. Without it, the model just confirms whatever you handed it looks fine.

2. “Don’t answer the question I asked. Diagnose where I’m actually stuck.”

I asked for better SMS copy for Apturo’s outreach. The real issue had nothing to do with the words. I hadn’t sent a single message in weeks. No copy was going to fix that. This prompt works because it gives the AI permission to look past your literal request and name the actual constraint, which is usually behavioral, not tactical. You have to be willing to hear “your copy is fine, you’re the problem” for this to be worth anything.

3. “Here’s my plan. Argue against it like a skeptical operator would, not like a cheerleader.”

I already run a BS detector on every guru post and offer that crosses my feed. I’d just never pointed that same instinct at my own plans. Most people ask AI to validate an idea and get back exactly that, validation. This prompt asks it to find the holes first: what assumption are you making that might be wrong, what’s the failure mode nobody’s mentioned, what would a competitor exploit. Cheap insurance before you spend money or time on something.

4. “Look across everything we’ve talked about and tell me the pattern you’re seeing that I’m not.”

This is the one that actually stings. It pointed out I’d parked the same two decisions more than once, both times waiting on the same unfinished thing to move first. I hadn’t put those together myself because I was living it one conversation at a time. AI that remembers context across sessions can see the repeat pattern you can’t, simply because you’re too close to it and it isn’t.

None of this is about smarter prompts for better content. It’s about giving the tool permission to push back instead of just agreeing with you.

The gap between 245K and $1M MRR was never going to close with more tactics. It closed when I stopped avoiding the send button.

Here’s why that matters. Once Apturo actually goes out and gets a contractors attention with a detailed report including competitor analysis, the close isn’t a pitch deck.

Within the report is a CTA. They call Sara AI where they hear our AI receptionist answer as if she works for their business, using their name, their services, their hours.

Most people are skeptical of AI voice agents right up until they hear it answer their own phone correctly.

The whole outbound system only exists to get a prospect to that one moment. Which means the bottleneck was never the demo.

It was getting enough doors to knock on, and I was the one not knocking.

As a founder we sometimes get caught up in the toys we build. We the difference is acknowledging the bottleneck and fix it.

When all was said and done this one tool increased our conversation rate by 27%. Why? Cause Sara did the selling… without selling.

What’s the thing you’re “still refining” that you actually just need to ship?